Device Information
Technical and regulatory information about the Skinpendium Clinical Decision Support System.
Rx Only — Prescription Device
Federal law restricts this device to sale by or on the order of a physician (21 CFR 801.109). Skinpendium is investigational, not FDA-cleared.
Device Classification
- Device Name
- Skinpendium Clinical Decision Support System
- Proposed Classification
- Class II
- Product Code
- QZS
- Regulation Number
- 21 CFR 878.1830
- Generic Name
- Software-aided adjunctive diagnostic device for use by physicians on lesions suspicious for skin cancer
Intended Use Statement
The Skinpendium device is indicated for use to evaluate dermoscopic images of skin lesions suggestive of melanoma, basal cell carcinoma, and/or squamous cell carcinoma in patients aged 40 and above to assist in the decision of biopsy or referral to a dermatologist.
The Skinpendium device analyzes and gives a probability of malignancy, likely classification of the lesion, and retrieves similar cases from a validated database of histopathology-confirmed/dermatologic-consensus lesions that the model was trained on to support clinical decision-making.
The device should be used in conjunction with the totality of clinically relevant information from the clinical assessment by physicians who are not dermatologists. The device should be used on lesions already assessed as suspicious for skin cancer and not as a screening tool. The device should not be used as the sole diagnostic criterion nor to confirm clinical diagnosis of skin cancer.
System Architecture
Skinpendium is deployed as a three-tier system: client applications, a cloud API, and the AI engine.
Client Applications
Mobile app (iOS and Android) for dermoscopic image capture, offline-first sync, and patient consent. An EMR plugin pulls Patient demographics and writes DiagnosticReport (AI result) plus ServiceRequest / Task (biopsy referral and follow-up) back to the EHR with server-side audit logging for every writeback.
Cloud API
FastAPI REST server handles inference requests and data synchronization. API key authentication (SHA-256 hashed, per-organization), HIPAA-compliant audit logging, rate limiting, and request validation. TLS 1.2+ for all communication.
AI Engine
A frozen vision foundation model extracts an embedding from each image. A binary malignancy head and a lesion-type classifier produce probabilities calibrated by isotonic regression, and a similarity index retrieves comparable confirmed cases.
Supported EHRs
- Epic
- Oracle Health (Cerner)
- SMART Health IT sandbox (development)
AI Pipeline
Each dermoscopic image is processed through a six-stage pipeline.
| Stage | Component | Output |
|---|---|---|
| 1. Feature extraction | Vision foundation model (frozen) | Image embedding |
| 2. Binary classification | Linear classification head | Malignant/benign probability |
| 3. Differential diagnosis | Lesion-type classifier | Probability for each lesion type |
| 4. Calibration | Isotonic regression | Calibrated probabilities |
| 5. Similar case retrieval | Similarity index | Most similar confirmed cases |
| 6. Clinical output | Aggregation layer | Biopsy/referral recommendation |
Technical Specifications
Model Architecture
- Backbone
- Vision foundation model (frozen)
- Outputs
- Malignancy probability, lesion type
- Calibration
- Isotonic regression
- Evidence
- Similar confirmed cases
Security & Compliance
- Fernet/AES-256 encryption at rest
- TLS 1.2+ for all API communication
- HIPAA Safe Harbor de-identification
- No PHI stored in audit logs
- API key authentication (SHA-256 hashed)